AI
Image to prompt
Reverse the recipe.
Get a prompt from any picture, and the original prompt itself when the image still carries it.
Prompt style
Plain sentences, for ChatGPT, Gemini, Flux and DALL·E.
If the image still holds the prompt it was made with, as Stable Diffusion images often do, you get that exact prompt. Otherwise the prompt is written from what's visible, so add the style or mood you want.
Describes a photo in a sentence or two. The first run takes a while.
Downloads once on first use, then runs offline. Your images are never uploaded.
How did this tool do?
Good to know
Where a prompt comes from, and how exact it can be
There are two very different answers to what prompt made an image. The first is the real one: Stable Diffusion apps such as Automatic1111, ComfyUI, InvokeAI and Fooocus write the prompt into the PNG they save, along with the negative prompt, model and settings. When a file still carries it, this tool shows it word for word, and nothing else could be more accurate.
Most images have lost that record, or never had it. ChatGPT, Gemini and Midjourney's website don't save the prompt in the file, and social networks and messaging apps strip metadata on upload. For those, a vision model looks at the picture and writes a description of the subject, setting, colours and light, which is then shaped into a prompt.
The three styles shape the same description differently. Natural keeps full sentences, which ChatGPT, Gemini, Flux and DALL·E follow well. Tags breaks it into short comma-separated phrases, the way Stable Diffusion prompts are usually written. Midjourney keeps the sentences and adds --ar with this image's own proportions, so a new version comes out the same shape.
A written prompt describes what is visible, not how it was made, so it won't name the model, the seed or an artist's style. Treat it as a starting point: run it, compare, then add the mood, lens or style words that the original seems to have had. The model downloads once and runs on your device, so the images you check are never uploaded.
Where you'll meet it
Recreate a look
A prompt to start from when you want something in the same spirit.
Recover your own settings
The exact prompt and negative prompt from your Stable Diffusion PNGs.
Learn prompt writing
See how a scene you like breaks down into words.
Guide
How to use it
- 01Drop your images, or paste one with Ctrl+V.
- 02Pick a prompt style: natural sentences, Stable Diffusion tags, or Midjourney with the aspect ratio.
- 03Tap Get prompt, then copy it or download every prompt as text.
Good to know
Questions, answered
Can it find the exact prompt an image was made with?
When the image still carries it, yes. Stable Diffusion apps such as Automatic1111, ComfyUI, InvokeAI and Fooocus save the prompt inside the PNG, and it's shown word for word. Images from ChatGPT, Midjourney's site or social media don't keep it, so a prompt is written from the picture instead.
Will the prompt recreate the same image?
Not exactly. A prompt describes the scene, but the model, seed and settings also shape the result. Expect something in the same spirit, then adjust the wording.
Which style should I pick?
Natural for ChatGPT, Gemini, Flux and DALL·E, which follow plain sentences. Tags for Stable Diffusion. Midjourney adds --ar so the new image keeps the same shape.
Is my image sent to a server?
No. The AI model itself is downloaded to your browser the first time you use the tool, then everything runs on your device — the image never leaves it. After that first download the tool works offline.